Mlucas v19.1 available

[kriesel]

aFor about half the stated price, and no ID theft cooperation required
https://www.ebay.com/itm/301323374960
and plugs into a standard US 120VAC 15A or 20A outlet. Variations are available for other nations' standards.
All these, including the original KillaWatt, obstruct both outlets of a duplex outlet.
Your local building supply or hardware store will gladly supply a cube tap or multi-outlet power strip to deal with that.

[ewmayer]

Originally Posted by bayanne

My new iMac will be arriving shortly, and I hope to compile and run MLucas v19.1 on it.

However where do I find the simple version on how to compile on ARM cpu?

Hi, Bayanne - tdulcet and danc2 have put together a very nice auto-install-and-tune script, but that is alas Linux-only. But a simple DIY command-line build for Arm-based iMac is straightforward, here the condensed instructions - this assumes you have the native Clang compiler installed, 'which clang' to check that. First download the v19.1 source tarball, which is linked at the Mlucas README page, unpack as detailed, then

cd ~/*19.1
mkdir obj
cd obj
clang -c -O3 -DUSE_ARM_V8_SIMD -DUSE_THREADS ../src/*.c >& build.log
grep error build.log
[Assuming above grep comes up empty]clang -o Mlucas *.o -lm -lpthread -lrt

Note Clang does not need explicit library-linkage of the math, pthread and realtime libraries, i.e. does not need the user to invoke -lm -lpthread -lrt at link time; if you include them and get a Clang linker error, try omitting them and relinking with just clang -o Mlucas *.o

To autotune for 8-core iMac M1, a hybrid of 4 'performance' and 4 'efficiency' cores: ./Mlucas -s m -iters 1000 -cpu 0:7

Assuming all that works, please post a copy of the resulting mlucas.cfg file containing optimal-FFT-parameters for your machine here. You can do assignments auto-management using the primenet.py script, a copy of which is in ~/*19.1/src . I suggest you create a run-directory at top level in your user-dir, copy the Mlucas binary, mlucas.cfg and primenet.py there, then run primenet.py as described at the README.

[bayanne]

Thanks, will let you know how I get on ...

[ewmayer]

How much RAM are you getting in your iMac? I really dislike the lack of RAM-expandability in the M1-based Apple offerings, as detailed in articles like this:

Theoretically, you can upgrade RAM & SSD on your M1 Mac mini, but you shouldn't | AppleInsider

Been busy last few months adding p-1 factoring support to the in-development v20 code - stage 2 really benefits from lots of RAM. Glad I always use humbler micro-PC-style systems among my optimization targets, making sure stage 2 performs at least halfway-decently with just 4-16GB RAM will also come in handy on M1 systems.

[bayanne]

How much RAM are you getting in your iMac?

As much as I was able to get, i.e. 16gb

[ewmayer]

Originally Posted by bayanne

As much as I was able to get, i.e. 16gb

For p-1 of current-wavefront (6M FFT) exponents, that should allow you to cut 20-25% work off the minimum-memory option for my stage 2 implementation, depending on how much of it you allow stage 2 to use: 8GB => 20%, 12GB => 25%. I'm working on the v20 stage 2 restart-from-interrupt code right now.

[bayanne]

Originally Posted by ewmayer

For p-1 of current-wavefront (6M FFT) exponents, that should allow you to cut 20-25% work off the minimum-memory option for my stage 2 implementation, depending on how much of it you allow stage 2 to use: 8GB => 20%, 12GB => 25%. I'm working on the v20 stage 2 restart-from-interrupt code right now.

Stuck at present, sent you a PM ...

[bayanne]

Originally Posted by bayanne

Stuck at present, sent you a PM ...

Now built, and crunching away, thanks for your help.

Is there a way of contacting Primenet with a progress report on the exponent?

[ewmayer]

Originally Posted by bayanne

Is there a way of contacting Primenet with a progress report on the exponent?

As I noted in my earlier e-mail, see the "Get exponents from Primenet" section of the README - it's linked in the Index at top of the page; please read it carefully, as it's your first go.

You'll need to copy the primenet.py script from your /src dir to whatever dir(s) you are planning to run instances from - for Apple M1, just a single instance using 1-8 threads (depending on how much you want to stress your CPU & cooling) seems a good compromise between throughput and ease-of-use.

[bayanne]

Originally Posted by ewmayer

As I noted in my earlier e-mail, see the "Get exponents from Primenet" section of the README - it's linked in the Index at top of the page; please read it carefully, as it's your first go.

You'll need to copy the primenet.py script from your /src dir to whatever dir(s) you are planning to run instances from - for Apple M1, just a single instance using 1-8 threads (depending on how much you want to stress your CPU & cooling) seems a good compromise between throughput and ease-of-use.

I probably did not express my request very clearly. I was hoping to update my Account Assignments Details which at present only show that my assignment was assigned on 2021-06-16 and that the last update was the same day. I have tried running primenet.py but it has not updated my progress on the exponent, or given the the estimated finish time.
Can this be done with primenet.py?

[M344587487]

Originally Posted by bayanne

Now built, and crunching away, thanks for your help.

Is there a way of contacting Primenet with a progress report on the exponent?

If you have a watt meter, it'd be interesting to know how much power the M1 chip draws under mlucas load. I've only just noticed Lorenzo's stats otherwise I'd be bugging you for that too, what a tasty thing to have missed.


Apple's chips are intriguing, almost to the point where they may break down my incredibly anti-Apple stance. In a year or two if a Linux port gets stable enough that mac OS can comfortably be nuked from orbit they may get some cash from me yet.